Trac #33100: use a more reliable abs() replacement in is_hermitian().

Some finite fields for which is_hermitian() otherwise works are
missing abs(). Here we use the square root of z*z.conjugate() as a
replacement, which appears to work in all of the cases we care about.
Note that this would probably do the wrong thing if you tried to use
a nonzero tolerance with an (exact) finite field.

src/sage/matrix/matrix0.pyx

@@ -4043,17 +4043,18 @@ cdef class Matrix(sage.structure.element.Matrix):
             for (i,j) in self.nonzero_positions():
                 d = self.get_unsafe(i,j) - s*self.get_unsafe(j,i).conjugate()
 
-                # avoid abs() which is missing for finite fields.
-                if d > tolerance or -d > tolerance:
+                # abs() support is spotty
+                if (d*d.conjugate()).sqrt() > tolerance:
                     self.cache(key, False)
                     return False
         else:
             for i in range(self._nrows):
                 for j in range(i+1):
                     d = (   self.get_unsafe(i,j)
                           - s*self.get_unsafe(j,i).conjugate() )
-                    # avoid abs() which is missing for finite fields.
-                    if d > tolerance or -d > tolerance:
+
+                    # abs() support is spotty
+                    if (d*d.conjugate()).sqrt() > tolerance:
                         self.cache(key, False)
                         return False